| 类别 | 参数 | 值 |
| 训练参数 | num_train_epoch | 5 |
| print_step | 10 | |
| batch_size | 64 | |
| batch_size_eval | 128 | |
| summary_step | 10 | |
| num_saved_per_epoch | 3 | |
| max_to_keep | 100 | |
| ALBERT | albert_small_zh_google | |
| 优化参数 | optim | adam |
| warmup_proportion | 0.1 | |
| use_tpu | None | |
| do_lower_case | TRUE | |
| learning_rate | 5.00E−05 | |
| TextCNN参数 | num_filters | 128 |
| filter_size | [2, 3, 4, 5, 6, 7] | |
| embedding_size | 384 | |
| keep_prob | 0.5 | |
| 其他参数 | sequence_length | 200 |
| weight_decay | 1.00E−06 | |
| seed | 666666 | |
| dropout | 0.3 |